Add overlooked overload information on torchlib functions

[BowenBao]

Stack from ghstack (oldest at bottom):

• Tool to modify torchlib overload names via libcst #920
• -> Add overlooked overload information on torchlib functions #919

Resolves most reported missing overloads from #865

[codecov]

Codecov Report

Merging #919 (1378db7) into main (241260f) will not change coverage.
The diff coverage is 100.00%.

@@           Coverage Diff           @@
##             main     #919   +/-   ##
=======================================
  Coverage   76.81%   76.81%           
=======================================
  Files         112      112           
  Lines       13547    13547           
  Branches     1377     1377           
=======================================
  Hits        10406    10406           
  Misses       2798     2798           
  Partials      343      343           

Files Changed	Coverage Δ
onnxscript/function_libs/torch_lib/ops/core.py	77.44% <100.00%> (ø)

[titaiwangms]

Are these all of them? I can start working on refactor get_aten_name in onnx dispatcher.

[BowenBao]

Are these all of them? I can start working on refactor get_aten_name in onnx dispatcher.

Updating with latest bench log.
There is aten::div.Tensor_Mode left, which should be unsupported, but somehow worked with fallback.

[justinchuby]

Maybe the rounding mode didn't affect things much?

[BowenBao]

Maybe the rounding mode didn't affect things much?

Maybe. Maybe type promotion does the right thing for it. Maybe rounding_mode was not provided so the signature matched, and behavior also matched for default true_divide.

I think if rounding_mode is explicitly set, export will fail, as it should.

[BowenBao]

@justinchuby do you know if CI / build_docs (windows-latest) (pull_request) failure is related?

[justinchuby]

No. It started to fail these two days